I think on loop is checked when you visit a new card, so that the new card has a chance to stop the loop, or start a new one. For example, if one card represents the player standing beside a waterfall, it would be more atmospheric if a loop of waterfall noise played whenever that card was visible. When the player visits that card, the loop should start immediately (even if some other loop was already part-way through playing) and when the player leaves that card, the loop should stop immediately (even if the waterfall noise was part-way through). Calling on loop on card transitions makes that kind of effect possible.
If your two songs are less than 10 seconds in total, you could combine them into one sound and loop that quite easily.
If your two songs are more than 10 seconds long in total, you might add a “last loop started at” field on the same card as the counter that switches between them. When you start playing a sound, update the counter, and set the “last loop started at” field to sys.ms (the current time in milliseconds). When on loop is called, it passes in the old sound - if the “last loop started at” time, plus the duration of the old sound in milliseconds (old.duration*1000, since the duration is in seconds), is nearly equal to sys.ms, then we’re probably looping because we got to the end of the sound. If the start time + the duration is greater than sys.ms, then the loop has been called early, probably because of a card change. If you return the original sound, it will continue playing it, rather than restarting it.